Railway Condition Monitoring and Fault Detection Based on YOLOv4

Abstract

Regular maintenance of the line is critical for safety in railway transportation, which constitutes an important part of the transportation system. Due to the high error rate of manual fault detection methods, non-contact fault detection methods have been developed. Railway switch state and level crossing faults are frequently encountered in the occurrence of train accidents. A new method based on YOLOv4 has been proposed for condition monitoring and fault detection in these rail sections. The YOLOv4 deep neural network was trained using four class label datasets consisting of real railway visual data. Model evaluated using test data. Performance evaluation was made using evaluation metrics. Experimental results showed that the model could detect correctly with 96.8% accuracy. © 2022 IEEE.